The minute’s defining feature was concentration, not congestion. Of 1,468 eligible transactions, address 0xcde6db…6c0e04 submitted 573—39.0% of the entire minute. Within those 573 records, 318 were classified as order cancellations and 254 as limit-order placements. Across the 1,453 transactions with enough evidence for behavior shares, cancellations accounted for 400 (27.5%) and placements for 361 (24.8%). Together, those two activities made up 761 of 1,453 behavior-classified rows.
The cadence reinforces the impression of a tightly automated maintenance loop: the median gap between transactions was 0 ms, the 90th-percentile gap was 220 ms, and activity occupied 59.817 seconds of the minute, with a longest idle period of only 538 ms. That is consistent with programmatic execution, but the address is only an on-chain identity; the evidence does not establish who operated it or whether the other addresses belonged to the same party.
What the dominant sequence looked like
Two late-minute transactions show the rhythm concretely. HY9cMGDL generated an owner proof and passed it to a DeepBook limit-order call. Its stored narrative records a 1,044.2 NS order at a resting price of 0.01542 SUI per NS, with no evidence that this was an execution price. Moments later, AqFiZZuu generated the same kind of proof and called cancellation for an identified order. The row records 1,051 NS at a 0.01532 SUI-per-NS resting price and explicitly distinguishes that resting size from anything traded.
This is order-book maintenance: DeepBook is documented as an on-chain central-limit order book, where placing and cancelling resting orders are distinct from realized exchange activity. [Sui’s DeepBook documentation](https://docs.sui.io/) provides that broader protocol context. These two rows support a cycle of placing and removing orders, not a claim that the address completed trades or earned anything.
The minute was not a system-wide failure event. There were 84 failed eligible transactions out of 1,468, or 5.7%, below the frozen corpus rate of 8.36%. Failures nevertheless clustered sharply: five stopping points accounted for 75 of 84 failures. para::repay_loan_base accounted for 21 (25.0%), flash_loan::return_flashloan_quote for 17 (20.2%), framework balance::split for 14 (16.7%), unrecorded sites for 13 (15.5%), and para::repay_loan_quote for 10 (11.9%).
4VrdVMCS is a useful exception to the dominant DeepBook rhythm. It opened short-lived flash credit, passed value through exchange-related calls touching SUI/CETUS and CETUS/USDC, then stopped at para::repay_loan_base with Move abort code 401. The transaction narrative says application changes were rolled back; the sender’s boundary movement was −0.00040664 SUI, with gas recorded separately. The reached exchange-related calls therefore show an attempted route, not a completed exchange.
The money is mixed, not a profit statement
Across the minute, sender-attributed SUI holding movements covered 538 transactions: 3,103.39660264 SUI in and 11,927.367562708 SUI out. Separate gas records covered 1,212 transactions, with 0.101304794 SUI in and 2.13043816 SUI out. These evidence streams are not additive, and there are no prices or cross-asset valuations here, so they do not establish profit or loss.
The cleanest reading of 19:28 is therefore a high-frequency order-maintenance burst led by one address, surrounded by a smaller set of attempted leveraged routes that repeatedly failed during repayment. The chain records the sequence and its boundaries; it does not identify the operator or prove that any order filled.
For a closer look, start with HY9cMGDL and AqFiZZuu, compare them with the failed route 4VrdVMCS, or follow the reusable pattern sui/cancel-order/deepbook-owner-proof/v1.